(622e) Robust Averaging Level Control

Surge tanks are widely used within the process industry to prevent upstream flow variations from upsetting downstream processes. The capacity of the tank is used to average out the effects of inlet flow variations thus attempting to keep the outlet flow smooth. The research area of averaging level control dates back to the late 1970s and since then a large variety of both linear and nonlinear controllers as well as innovative control structures have been proposed.

Typical earlier approaches to averaging level control assumes a step change to the inlet flow and given a maximum permitted deviation of tank level the optimal tuning of the controller is given. In addition the controller needs to be robust towards future inlet flow variations, which is achieved by returning the tank level to 50%. The weakness with this robustness strategy is exposed when two subsequent step changes in, e.g., the positive direction occurs within a short time interval. The first step change results in an increased tank level and the second step then has to be more or less directly transferred to the outlet flow to avoid tank overflow, since less tank capacity is available. One way to remedy this type of situation is of course by tighter tuning, which yields worse flow filtering for all types of inlet flow changes. Furthermore it is also difficult to ensure that a similar situation will not occur again in the future.

We have taken a more direct approach to achieve good flow filtering and still be robust towards future inlet flow changes namely closed loop robust MPC. Given that the user specifies the minimum and maximum inlet flow as well as the permitted minimum and maximum tank level the robust MPC controller achieves optimal flow filtering while accounting for future variations of the inlet flow and guaranteeing no tank level violation. The minimum and maximum inlet flow can typically be estimated from historical data, perhaps adding some margin for additional robustness, while the minimum and maximum permitted tank level are at the hands of the user. The computational burden of robust MPC is admittedly rather high but the conclusions that can be drawn from this implementation is on the other hand very relevant and worthwhile. The robust MPC controller behaves significantly different from earlier proposed controllers in that it does not return the tank level to 50% following an inlet flow step change. Instead the new steady state tank level depends on the level of the inlet flow, e.g., 0% inlet flow yields the user-defined minimum allowed tank level. This type of level control, where the tank level is not returned to 50%, but rather is a function of the inlet flow has traditionally only been associated with proportional only controllers. That the same behavior is observed using robust closed loop MPC we take as an incentive to further explore the concept. One straightforward extension is to let a feed-forward signal from the measurable or estimated inlet flow be the set-point of the level controller.